Deploybase CLIGPU + LLM pricing in your terminal.
dbdeploybase

LLM PRICING

Command R+ Pricing: Compare Costs Across All API Providers

Comprehensive analysis of Cohere Command R+ pricing across all API providers, cost comparison with Claude and GPT-4, and usage optimization strategies as of March 2026.

By Deploybase · July 24, 2025

Contents

Cohere Command R+ Overview

Command R+ represents Cohere's latest open-weights large language model. Released in March 2024, the 104B parameter model competes directly with Claude Opus and GPT-4 on reasoning and coding tasks.

Key characteristics:

  • Parameter count: 104B (open-weights, can be self-hosted)
  • Context window: 128K tokens
  • Training data cutoff: February 2024
  • Instruction-tuned for dialogue and task completion
  • Multilingual support across 100+ languages

Cohere positions Command R+ as cost-effective alternative to proprietary models. Open-weights variant enables self-hosting on consumer cloud GPUs for further cost reduction.

Official Pricing

Cohere API (cohere.AI)

Cohere charges per-token pricing with distinct rates for input and output:

  • Input tokens: $0.0025 per 1K tokens ($2.50 per million)
  • Output tokens: $0.010 per 1K tokens ($10.00 per million)

Example cost for 10K input tokens + 2K output tokens: (10 * $0.0025) + (2 * $0.010) = $0.025 + $0.020 = $0.045 per request

Volume discounts apply:

Monthly VolumeInput RateOutput Rate
< 1M tokens$0.0025$0.010
1M - 10M$0.0020$0.008
10M - 100M$0.0015$0.006
100M+$0.0010$0.004

At 100M+ tokens monthly, effective rates drop 60% to $0.0010/$0.004. Equivalent to $1.00 per million input tokens, $4.00 per million output tokens.

Trial Period

New accounts receive $500 free credit. Sufficient for 8-16 million tokens depending on input/output ratio. No credit card required for trial period.

Third-Party Provider Rates

Multiple platforms resell Cohere API access with varying markups:

OpenRouter

OpenRouter integrates Cohere Command R+ with 5-10% markup:

  • Input: $0.002625 per 1K tokens (5% markup on $2.50/M)
  • Output: $0.01050 per 1K tokens (5% markup on $10.00/M)

Advantage: Unified billing across all providers. Disadvantage: Small cost premium.

Together AI

Together AI hosts Cohere models on their infrastructure with different pricing:

  • Input: $0.0025 per 1K tokens
  • Output: $0.0105 per 1K tokens

Similar to Cohere pricing but Together applies different token counting methodology, sometimes resulting in 5-8% higher effective costs.

Mistral AI Platform

Mistral partners with Cohere to offer Command R+ access:

  • Input: $0.00208 per 1K tokens (17% discount on $2.50/M)
  • Output: $0.00830 per 1K tokens (17% discount on $10.00/M)

Mistral's lower rates reflect infrastructure partnership and volume commitments.

Replicate

Replicate offers Command R+ through their platform:

  • Input: $0.003325 per 1K tokens (33% markup on $2.50/M)
  • Output: $0.01330 per 1K tokens (33% markup on $10.00/M)

Higher cost reflects Replicate's management and monitoring services. Suitable for teams prioritizing operational simplicity over cost optimization.

Cost Comparison

Comparing three models at 1M input tokens + 500K output tokens monthly (typical usage):

Cohere Command R+ (direct API)

  • Input: 1,000 * $0.0025 = $2.50
  • Output: 500 * $0.010 = $5.00
  • Total: $7.50

OpenAI GPT-4 Turbo (see OpenAI pricing for current rates)

  • Input: 1,000 * $0.01 = $10.00
  • Output: 500 * $0.03 = $15.00
  • Total: $25.00

Anthropic Claude Opus 4.6

  • Input: 1,000 * $0.005 = $5.00
  • Output: 500 * $0.025 = $12.50
  • Total: $17.50

Groq Mixtral (API)

  • Input: 1,000 * $0.0005 = $0.50
  • Output: 500 * $0.0008 = $0.40
  • Total: $0.90

Cohere Command R+ saves 70% versus GPT-4 and 57% versus Claude Opus 4.6. However, Groq's edge models offer 88% cost reduction. Quality differences require benchmarking specific use cases.

See the complete LLM API pricing comparison for full analysis across all providers.

Usage Optimization

Prompt Caching Strategy

While Cohere doesn't offer native prompt caching like Claude, alternative strategies reduce input token consumption:

System prompt optimization: Minimize system prompt length. Each request repeats system tokens. Reducing system prompt from 2K to 500 tokens saves $1.50 per 1M requests.

Retrieval-augmented generation: Embed documents client-side, only send top-K relevant chunks to API. Reduces input tokens by 60-80% versus sending full documents.

Batch Processing

Cohere offers batch API with 20% discount:

from cohere.client import Client

client = Client(api_key="YOUR_API_KEY")

requests = [
    {"text": "Explain photosynthesis", "max_tokens": 200},
    {"text": "What is quantum entanglement?", "max_tokens": 200},
]

batch_response = client.create_batch(requests)

Batch processing queues requests for non-real-time processing. Cohere processes during off-peak hours and applies 20% discount. Suitable for email summaries, content moderation, and data labeling.

Model Tier Selection

Cohere also offers Command R (non-Plus variant) at lower cost:

Command R pricing:

  • Input: $0.00015 per 1K tokens ($0.15 per million, 94% cheaper than R+)
  • Output: $0.00060 per 1K tokens ($0.60 per million, 94% cheaper than R+)

Command R trades capability for cost. Benchmarks show Command R performs adequately on simpler tasks (summarization, classification) but struggles with complex reasoning.

Self-Hosting Open Weights Model

Deploy Command R+ locally on cloud GPU for ultimate cost control. Using RunPod H100:

  • H100 rental: $2.69/hour
  • Assumptions: 3 requests/second, 8-hour operating window
  • Cost per million tokens: $0.11 (input and output combined)

This undercuts all API providers by 92%. Suitable for production workloads exceeding 500M tokens monthly.

See the guide on small open-source LLMs on consumer GPUs for deployment details.

FAQ

Q: How does Command R+ quality compare to GPT-4 and Claude Opus 4.6? Command R+ matches GPT-4 on coding and reasoning benchmarks but underperforms Claude Opus 4.6 on complex multi-step reasoning. Differences are task-specific. Benchmark against your specific use cases before committing.

Q: What's the cost per token effectively at scale? Direct Cohere API at 100M+ tokens monthly: $0.0010 input / $0.004 output per 1K tokens ($1.00/$4.00 per million). With typical 1:2 input:output ratio, blended cost of $0.003 per 1K tokens. Compare to GPT-4's $0.013 blended rate per 1K tokens.

Q: Should I self-host Command R+ to save costs? Only if monthly token volume exceeds 500M tokens. Below that threshold, API pricing cheaper due to GPU idle time. Above 500M, self-hosting saves 85-90% versus API costs.

Q: Which third-party provider offers best Command R+ pricing? Mistral AI Platform with 17% discount below Cohere's rates. OpenRouter offers convenience of unified billing but small 5% markup. Choose based on integration simplicity versus cost priority.

Q: Does Cohere offer SLA guarantees? 99.9% uptime SLA for production instances. Production tier (custom pricing) provides 99.99% SLA with dedicated support. Standard tier appropriate for non-critical applications.

Q: Can I use Command R+ for fine-tuning? Yes. Cohere fine-tuning available for Command R+ with per-token training costs. Typically $0.50-$2.00 per million training tokens depending on adapter configuration. Contact Cohere sales for custom pricing.

Sources

  • Cohere Official Pricing Documentation (March 2026)
  • OpenRouter API Pricing
  • Together AI Pricing Documentation
  • Mistral AI Platform Pricing
  • Replicate Service Pricing
  • Industry Model Benchmark Aggregation